Who killed the Intel microprocessor?
asymco.com
asymco.com
a: licensing cost / production cost
b: Power consumption
If you think that intel wont have both of those factors down to ARM levels shortly, you dont know intel very well. They are very competitive and dont like to get beat.
The graphically intensive games of tomorrow are not optimized for many cores (still) and will need the raw power of a higher clock. This is especially true for strategy games like starcraft that make heavy use of the CPU.
As time moves on, I do imagine the ability to run x86 operating systems / applications will be worth a lot as well. Yes, there may be virtualization as a option but I am not optimistic in the performance of such a feature.
Back to your question. When you asked what the motivation will be to switch to x86 in 3-4 years, I think the answer is: Even in 3-4 years, the amount of developers familiar with x86 and the sheer amount of software already available for it will trump the offerings of ARM.
What has Intel been doing instead to gain performance? Architectural optimizations, more cores, etc. The same game ARM plays.
I'm not holding my breath for 5GHz chips. It might happen, but the exponential growability of raw clock speed is well and truly over. We ran into that brick wall in 2003. We're slowly crawling the clock speeds up, but to rely on it as a competitive advantage is IMHO lunacy.
Also, I have to seriously challenge the notion that games to tomorrow are not optimized for multiple cores. This simply isn't true, particularly as more and more developers fall back onto common middleware like the Unreal engine, and almost no games are exclusively developed for a single platform (The Xbox 360 and PS3 both require extensive parallel programming support to really harness its capabilities).
Games coming out today already can take advantage of multiple cores, albeit there's a lot of work that can be done on that front still. Games of tomorrow are certainly not architecturally single-core.
> "Even in 3-4 years, the amount of developers familiar with x86 and the sheer amount of software already available for it will trump the offerings of ARM."
Who is actually "familiar with x86"? Most programmers do not work at the microarchitecture level of the chain. Most of us write in higher-level languages and develop apps directly. Sure, in 3-4 years the number of low-level developers who understand x86 better than ARM will still hold, but for the majority of the programming world (the ones that never touch architecture-related issues) it makes no difference. It's a different set of flags to gcc and a few gotchas you have to learn.
Game developers develop mainly for... PowerPC :) -> Nintendo Wii, Xbox 360, PS3 are all PPC architectures.
Apart from certain MMO's - console video games are doing way more sales than PC.
That's not true. Console video games have 5+ years of experience dealing with multiple cpus, and some even with different ones communicating through DMA (ps3).
There are physics, sound, graphics, even data compression libraries optimized to use all CPUs there are.
The problem is the PC games where you don't really know on which machine your game would land on - dual core, or quad core - what kind of memory latency, etc.
If you are making an MMO you would shoot for more widespread PC, even laptop/notebooks.
If you are shooting for high-quality FPS game, not caring about sales much then you can require high-level machines (but you won't get much money back).
Or provide options for graphical selection.
What do you base your lack of doubt on? Any hard info you can point to?
Costs have been cut in every place possible in order to maximize shipment volume to make consumer-level smartphones possible on thin margins.
Intel's Atom x86 is Intel's lowest margin x86 CPU. Accordingly, it is built on years-old hand-me-down fabrication process, with very little investment put in to power saving technology. Despite this, for years now it only consumes a watt or so per core, with the cut-cost platform (chipset) surrounding it in the average laptop, consuming vastly more.
Intel already has a 32nm System-on-a-chip platform developed, simply waiting for the right time that it makes sense to ramp up volume.
And so, the thing that can kill Intel, is that performance from a budget-priced CPU out of a Taiwan fab is all that consumers want or need any more in any device.
Wrong.
A new Mac Pro has about twice the single-threaded performance, according to the Geekbench scores. Compare this: http://browse.geekbench.ca/geekbench2/view/339104 to this: http://browse.geekbench.ca/geekbench2/view/281328
OK, memory bandwidth is a lot better, about 5 times as fast, but otherwise the performance gains are pretty shoddy. Processors have not been getting much better at single-threaded work.
Keep in mind, this is comparing a cheap laptop to an expensive workstation.
If a cheap fab is 5 years behind Intel, it's already breathing down Intel's neck.
If you are speaking about server applications: ARM doesnt really have a shot here. The sheer computing power differences are too much to have users downgrade to ARM. Yes, you can save XXX on your power bill but most people rent servers, not colo. Additionally, you will likely need 3-4x machines to equal the power of the x86 machine you are replacing, negating any power benefit.
Regarding consumer machines: No, power usage does not matter. Except in the rare cases ("think green" crazies), Nobody checks the power usage of a computer they are buying at best buy. They check processor speed, memory and harddrive space. (and price)
Regarding price: Consumer machines are heavily subsidized by pre installed software, volume discounts to both distributor and manufacturer (ie- HP and best buy ) and other factors. The cost for desktop machines, is already very near what ARM could be (if you need performance- We are no talking about 100$ netbooks here). People pay a premium for quality and speed. Thats the reason MBP's cost 25-33% more than a comparable dell/lenovo.
In regards to mobile: Battery life is important, but I think most people forget to check this statistic before buying a phone. Regardless, as intel gets more and more into this area they should be able to match/beat the offerings of ARM. Let not forget that ARM isnt all that amazing though. Just look at the power usage on a EVO for instance. ( http://techcrunch.com/2010/06/09/dont-buy-the-android-evo-it...)
Anyways: All my comments above hold, even considering that intels power performance isnt near what it needs to be yet. As it gets better and they enter mobile it will be taken more seriously.
The guy who is in charge of datacenter power efficiency at Amazon is super-enthusiastic about ARM servers:
http://perspectives.mvdirona.com/2010/11/20/VeryLowCostLowPo...
James writes a lot about power efficiency in server computing, and his blog is a must-read for anyone who wants to have a qualified, researched opinion on the subject of power management on the server.
Technology is not the issue here. Technology already can give us a quadcore x86 CPU in a smartphone - the question is just how much are you willing to pay for it.
As long as Intel can keep making chips 5x bigger and sell them for 20x the profit, they will. Consumer demand doesn't appear to be slowing for this; embedded CPUs are simply opening a new market for portable, low power devices with acceptable performance.
Embedded system-on-a-chip margins are very low, and thus the shipment volume of smart phone and tablet CPU shipments has to be very, VERY high before it can begin to bring Intel the sort of profit margins they already enjoy for x86 consumer parts.
Now, if you're saying, "yeah, but not for desktops", I can agree with you. Desktops are going to become less and less powerful as the average computer user becomes more and more computer illiterate ("the Facebook dumb terminal", I'll call it), and ARM may help keep the cost down. But I'm probably not going to be running Emacs on an ARM box anytime soon. (Hell, I even use x86 for my router.)
AMD P/E: 5.46
Intel P/E: 11.12
ARM P/E: 86.94
If anything, it's the current mobile model that will go obsolete at some point -- simply because it's costly, both in terms of customizing SoCs, taping out small batches of silicon and customizing software for it. Mobile will shift to commodity components at some point.
Selling tens of millions of a single hardware model doesn't count as commodity?
Even discounting that, ARM's business model doesn't preclude the emergence of 'commodity' ARM processors. Maybe such a shift might play to Intel's strengths, but then ARM has total domination of the mobile CPU market in their favour.
I have no idea where this market will go in say 10 year's time, but right here, right now ARM have a lot going for them and there are no short or medium term reasons to suspect that's likely to change.
Most ARM CPUs are manufactured not by ARM itself but by Samsung, Qualcom, Texas Instruments and the like. These guys certainly do have all the requisite capacity as they provide the silicon for all of the world's mobile phones.
Granted, the Big Gorilla is faster to shrink the processes, because they have no problem dropping a few billions on a brand-new fab, but what good did that do? On both supposed benefits of the smaller process (cost and power drain / performance) ARM CPUs are still winning.
This whole situation is a replay of the Pentium 4 Megahurtz madness. Back then Intel was pushing raw performance disregarding power consumption, then they hit the heat dissipation wall, AMD snuck in from behind and punished them in a major way with Athlon-64. Intel regrouped with Core* series, won back the market share, pushed AMD back into oblivion, and relaxed about the whole thing. Now they are doing the same thing - prioritizing performance at the cost of power drain, and this time it will be ARM who will punish them. It's not clear how long it will take Intel to change their ways this time. The last time they still had Pentium 3 / Pentium M group of engineers to lean on, and that they did rather well. This time they are not so lucky - they have actually sold off their embedded CPU division a few years back, and now I doubt whether there is anyone left inside Intel to insist on a much needed pivot.
[EDITED for typos]
Not quite. With the exception of Samsung, those companies design ARM-based systems-on-a-chip, but do not manufacture them. They take an ARM core, perhaps modify it, add on a bunch of other stuff like graphics accelerators, memory controllers, ethernet MACs, and so on. Then once the design checks out, they send it to another company, like TSMC, UMC, or GlobalFoundries, that actually does the manufacturing.
The capital costs of building each new generation of chip fabs have grown tremendously, which led to this centralization and commoditization of chip manufacturing. Intel is one of the few holdouts, and they're usually a bit ahead of everybody else, which is one of their big competitive advantages.
I'd be really surprised if Intel isn't looking pretty good come this time next year.
Their best-in-industry process is actually sort of a problem for them, because they need those billions they dropped to translate into high margins. Which means they need their fabs to be making Xeons, not Atoms.
Or to put it another way: it isn't clear that mobile CPUs can pay for expensive fabs, which means the cheaper fabs in Taiwan have an advantage. Intel more or less admitted this when they ported the Atom to TSMC's process.
But then the question becomes: is an Atom made at TSMC better than an ARM made at TSMC?
And I didn't know about Atom/TSMC deal, so thanks for pointing that out: http://arstechnica.com/hardware/news/2009/03/atom-cant-feed-...
Quite humorous really.
ARM's business model is well matched to their market (mobile phone designers). Apple is selling to a different market (mobile phone users) with different requirements and dynamics, and so a different business model is appropriate.